diff --git a/README.md b/README.md
index a972776..6edcf98 100644
--- a/README.md
+++ b/README.md
@@ -8,10 +8,10 @@
 
 A single binary that:
 
-- **Accepts a prompt and returns a result** (Mode A, single-turn): `amplifier-agent run "your prompt"`
-- **Speaks JSON-RPC over stdio for multi-turn conversations** (Mode B): `amplifier-agent run --stdio`
+- **Accepts a prompt and returns a result** (single-turn): `amplifier-agent run "your prompt"`
+- **Emits one JSON envelope on stdout per invocation** — wrappers spawn one process per turn and pass `--session-id` for continuity
 
-It is *not* a server, daemon, or long-lived service. Each invocation is a fresh process that exits when its caller closes stdin or sends `agent/shutdown`.
+It is *not* a server, daemon, or long-lived service. Each invocation is a fresh process that runs one turn and exits. Multi-turn conversations are managed at the wrapper or session-ID layer — not inside a persistent process.
 
 The engine library inside (`amplifier_agent_lib`) is transport-free Python that any Python app can also embed in-process — no subprocess needed.
 
@@ -19,19 +19,24 @@ The engine library inside (`amplifier_agent_lib`) is transport-free Python that
 
 Existing AI agent infrastructure assumes you're building a chat product. `amplifier-agent` is the opposite: it's an *engine you point other software at*. The CLI is the universal adapter — wherever you can shell out, you can use Amplifier.
 
-The wire protocol intentionally mirrors [MCP](https://modelcontextprotocol.io/) (JSON-RPC over stdio, server-initiated bidirectional requests, capability negotiation) so existing host clients can integrate with minimal new infrastructure.
+The Mode A wire protocol is intentionally simple: the engine takes a single invocation (argv + env), runs one turn, and writes one JSON result envelope to stdout. Wrapper SDKs (TypeScript and Python) handle spawning, result parsing, and session continuity on top.
 
 ## Install
 
+The Python engine is not yet published to a package registry, but `uv tool install` installs it directly from git:
+
 ```bash
-uv tool install amplifier-agent
+uv tool install git+https://github.com/microsoft/amplifier-agent
 amplifier-agent doctor       # verify environment
 ```
 
-Other install methods:
+Pin to a specific engine release by appending a tag:
+
+```bash
+uv tool install git+https://github.com/microsoft/amplifier-agent@engine-v0.3.0
+```
 
-- `pipx install amplifier-agent`
-- From source: `git clone … && cd amplifier-agent && uv sync && uv tool install -e .`
+Engine and wrapper releases are tagged separately (`engine-v0.3.0`, `wrapper-v0.4.0`). For local development against a checkout, `git clone` the repo and run `uv tool install -e .` from inside it.
 
 First-run will prepare the built-in bundle and cache it to `$XDG_CACHE_HOME/amplifier-agent/`. Subsequent invocations skip this step.
 
@@ -49,12 +54,7 @@ Run a one-shot turn:
 amplifier-agent run "Summarize the README of github.com/microsoft/amplifier"
 ```
 
-Or wire it into a host as a JSON-RPC subprocess:
-
-```bash
-amplifier-agent run --stdio
-# Then write JSON-RPC requests to stdin, read events from stdout
-```
+The TypeScript and Python wrapper SDKs handle subprocess management automatically. See `wrappers/typescript/` (`amplifier-agent-ts` on npm) and `wrappers/python/` for ready-to-use clients.
 
 ## Provider configuration
 
@@ -71,10 +71,9 @@ Override with `--provider <name>`. No `settings.yaml` to maintain.
 
 | Mode | Invocation | Caller | Lifecycle |
 |---|---|---|---|
-| **A** (single-turn) | `amplifier-agent run "prompt"` | Shell scripts, OpenClaw skills, ad-hoc CLI use | Spawn → init → one turn → exit |
-| **B** (multi-turn stdio) | `amplifier-agent run --stdio` | Wrapper SDKs, conversational host adapters | Spawn → init → many turns → exit on EOF |
+| **A** (single-turn) | `amplifier-agent run "prompt"` | Shell scripts, wrapper SDKs, host adapters, ad-hoc CLI use | Spawn → one turn → JSON envelope on stdout → exit |
 
-Both modes share the same engine; the only difference is who drives the I/O loop.
+Multi-turn conversations are built by the caller: spawn one process per turn, passing `--session-id` for continuity. A persistent stdio JSON-RPC mode (Mode B) was originally designed but was superseded by the Mode A subprocess driver in PR #8 (see `docs/designs/2026-05-24-aaa-v2-mode-a-pivot-amendment.md`).
 
 ## Session continuity
 
@@ -95,6 +94,8 @@ Sessions are persisted as transcript JSONL in `$XDG_STATE_HOME/amplifier-agent/s
 
 ```bash
 amplifier-agent doctor              # Diagnose env, providers, paths, bundle cache
+amplifier-agent prepare             # Pre-warm bundle cache (run once after install)
+amplifier-agent verify              # Verify install integrity
 amplifier-agent config show         # Print resolved config with source annotations
 amplifier-agent cache clear         # Invalidate the prepared-bundle cache
 amplifier-agent --version           # Print version
@@ -102,31 +103,86 @@ amplifier-agent --version           # Print version
 
 ## Approval flow
 
-Some tools (file writes, command execution) request approval before acting. In single-turn mode:
+Some tools (file writes, command execution) request approval before acting:
 
-- **Interactive terminal**: prompted on stderr; respond `y` / `N` / `c`
+- **Interactive terminal**: prompted on stderr; respond `y` to approve, anything else to decline
 - **Non-interactive (CI, pipe, background)**: denied by default
 - **Override**: `-y` accepts all, `-n` denies all (apt-style)
 
-In stdio mode, approval flows over the wire as `approval/request` server-initiated JSON-RPC requests. Wrapper SDKs implement the host-side handler (callback, message-back, email, or anything else creative — adapter's choice).
+Wrapper SDKs can install their own approval handler (callback, message-back, email, or anything else creative — adapter's choice) via the `ApprovalSystem` protocol point on the engine library.
 
 ## Embedding in your own Python host
 
 Skip the CLI entirely if your host is Python:
 
 ```python
-from amplifier_agent_lib import Engine
+import sys
+from amplifier_agent_lib import __version__
+from amplifier_agent_lib._runtime import make_turn_handler
+from amplifier_agent_lib.bundle.cache import load_and_prepare_cached
+from amplifier_agent_lib.engine import Engine
+from amplifier_agent_lib.protocol import PROTOCOL_VERSION
 from amplifier_agent_lib.protocol_points.defaults_cli import CliApprovalSystem, CliDisplaySystem
 
-engine = await Engine.boot(
-    approval_system=CliApprovalSystem(mode="auto"),
-    display_system=CliDisplaySystem(verbosity="normal"),
+prepared = await load_and_prepare_cached(aaa_version=__version__)
+handler = make_turn_handler(prepared, cwd=None, is_resumed=False, mcp_config_path=None)
+
+engine = Engine(
+    turn_handler=handler,
+    protocol_points={
+        "approval": CliApprovalSystem(mode="no"),
+        "display": CliDisplaySystem(verbosity="normal", stream=sys.stderr),
+    },
 )
-result = await engine.submit_turn(prompt="Hello!", session_id="my-session")
+await engine.boot({
+    "protocolVersion": PROTOCOL_VERSION,
+    "clientInfo": {"name": "my-host", "version": "0.1.0"},
+    "capabilities": {},
+    "sessionId": "my-session",
+    "resume": False,
+})
+result = await engine.submit_turn({
+    "sessionId": "my-session",
+    "turnId": "turn-1",
+    "prompt": "Hello!",
+})
 await engine.shutdown()
 ```
 
-See `src/amplifier_agent_lib/` for the full library surface.
+`Engine.boot()` is an instance method that takes a params dict. The constructor requires both `turn_handler` (built from a `PreparedBundle`) and the `protocol_points` dict. See `src/amplifier_agent_lib/` for the full library surface.
+
+## TypeScript / Node.js SDK
+
+For Node.js and TypeScript hosts, use the `amplifier-agent-ts` npm package. It is a thin process supervisor that spawns the Python `amplifier-agent` CLI per turn (Mode A) and exposes a typed async API — all inference, tool execution, and session state live in the Python engine.
+
+You need **both** packages installed: the npm SDK *and* the Python engine (see [Install](#install) above — the Python CLI must be on `PATH`).
+
+```bash
+npm install amplifier-agent-ts
+```
+
+```typescript
+import { spawnAgent, AaaError } from 'amplifier-agent-ts';
+import { randomUUID } from 'node:crypto';
+
+const session = await spawnAgent({
+  lifecycle: 'one-shot',
+  sessionId: randomUUID(),
+});
+
+try {
+  const result = await session.submit({ prompt: 'Hello, agent.' });
+  console.log(result.reply);
+} catch (err) {
+  if (err instanceof AaaError) {
+    console.error(`[${err.code}] ${err.message}`);
+  } else {
+    throw err;
+  }
+}
+```
+
+Requires Node.js ≥ 20. Zero npm runtime dependencies. Full API surface in [`wrappers/typescript/README.md`](wrappers/typescript/README.md) and the type definitions at `wrappers/typescript/dist/index.d.ts`. The Python sibling SDK lives at [`wrappers/python/`](wrappers/python/).
 
 ## Architecture at a glance
 
@@ -138,7 +194,7 @@ Host Application                              ← your code
 Adapter (host-specific glue)                  ← per-host integration
     ↓
 Language Wrapper (TypeScript or Python)       ← typed SDK
-    ↓ JSON-RPC over stdio (or in-process)
+    ↓ subprocess (argv in / JSON envelope out, or in-process)
 amplifier-agent CLI                           ← this repo
     ↓ (in-process)
 amplifier_agent_lib (engine library)          ← this repo
@@ -148,22 +204,46 @@ Amplifier Kernel (amplifier-core, amplifier-foundation)
 
 The CLI binary (`amplifier-agent`) is a thin I/O adapter on top of `amplifier_agent_lib`. The library is transport-free — Python hosts can skip the subprocess entirely.
 
-## Wire protocol (Mode B)
+## Wire protocol (Mode A)
+
+Protocol version: **`0.2.0`** (defined in `src/amplifier_agent_lib/protocol/methods.py`; breaking changes bump this). Wrappers must pass `--protocol-version 0.2.0` — version mismatches return a `protocol_version_mismatch` error and exit non-zero rather than silently misbehave.
 
-Mode B speaks JSON-RPC 2.0 over newline-delimited stdin/stdout:
+Mode A uses a single subprocess invocation per turn. The wrapper passes flags as argv; the engine writes one JSON envelope line to stdout on completion.
 
-| Method | Direction | Purpose |
+**Input (selected argv flags):**
+
+| Flag | Type | Purpose |
 |---|---|---|
-| `agent/initialize` | Host → Agent | Capability negotiation |
-| `session/create` | Host → Agent | Open a new session |
-| `turn/submit` | Host → Agent | Submit a turn; agent streams notifications + returns result |
-| `turn/cancel` | Host → Agent | Cancel an in-flight turn |
-| `session/end` | Host → Agent | Close session and persist state |
-| `agent/shutdown` | Host → Agent | Graceful exit |
-| `approval/request` | Agent → Host | Request approval for a sensitive action |
-| `notifications/*` | Agent → Host | Streaming events (`result/delta`, `result/final`, `tool/started`, `tool/completed`, `progress`, `thinking/*`, `usage`, `error`) |
+| `PROMPT` | positional | The turn prompt |
+| `--session-id` | str | Session ID for continuity |
+| `--resume` | flag | Resume from saved transcript |
+| `--fresh` | flag | Discard saved state and start over |
+| `--protocol-version` | str | Wrapper's pinned protocol version; engine validates match |
+| `--mcp-config-path` | path | Path to MCP server config JSON (written by the wrapper) |
+| `--host-capabilities` | JSON | Host capability advertisement |
+| `-y` / `-n` | flag | Auto-approve / auto-deny all approval requests |
+| `--output` | text \| json | `json` (default) emits full envelope; `text` emits the reply only |
+
+**Output (stdout, single JSON line):**
+
+```json
+{
+  "protocolVersion": "0.2.0",
+  "sessionId": "...",
+  "turnId": "turn-1",
+  "reply": "...",
+  "error": null,
+  "metadata": {
+    "tokensIn": 0, "tokensOut": 0, "durationMs": 0,
+    "bundleDigest": "...", "engineVersion": "...",
+    "protocolVersion": "0.2.0", "correlationId": "..."
+  }
+}
+```
+
+Diagnostic events (tool calls, thinking, progress) go to **stderr** only — stdout is reserved for the single envelope so callers can parse it with `JSON.parse(line)` without filtering.
 
-Notifications are one-way (no `id`). Server-initiated requests use the same ID-correlation as host-initiated ones, just in reverse.
+The TypeScript and Python wrapper SDKs (`wrappers/typescript/`, `wrappers/python/`) handle all of this: they spawn `amplifier-agent run`, write the MCP config tmpfile, parse the envelope, and expose a typed async API.
 
 ## Related repositories
 
@@ -176,13 +256,21 @@ Notifications are one-way (no `id`). Server-initiated requests use the same ID-c
 
 ## Status
 
-Phase 1 — Layer 4 (engine library + CLI) ships in this release. Roadmap:
+Current versions: engine `0.3.0`, TypeScript wrapper `amplifier-agent-ts@0.4.0`, Python wrapper `amplifier_agent_client` (see `wrappers/python/`). Wire protocol: `0.2.0`.
+
+**Shipped:**
+
+- L4 — Engine library + CLI (Mode A)
+- L3 — TypeScript + Python wrapper SDKs (`amplifier-agent-ts` on npm, `amplifier_agent_client` in `wrappers/python/`)
+- Protocol schemas + cross-language conformance test suite (`wrappers/conformance/`)
+- Path-based MCP config delivery (`--mcp-config-path`)
+
+**In progress / next:**
 
-- L3 language wrappers (TypeScript + Python SDKs) — designed, implementation next
-- L2 host adapters (NanoClaw, Paperclip) — designed, implementation after L3
-- Install paths, container packaging, full execution plan — deferred sections of the design checkpoint
+- L2 — Host adapters (NanoClaw, Paperclip) — see `docs/designs/2026-05-22-aaa-v2-amplifier-agent-nc-provider.md`
+- Container packaging, install-path finalization
 
-See [`docs/designs/`](docs/designs/) and the [pull requests](https://github.com/microsoft/amplifier-agent/pulls) for design history and roadmap.
+See [`docs/designs/`](docs/designs/) and the [pull requests](https://github.com/microsoft/amplifier-agent/pulls) for design history and roadmap. The Mode A pivot is captured in [`docs/designs/2026-05-24-aaa-v2-mode-a-pivot-amendment.md`](docs/designs/2026-05-24-aaa-v2-mode-a-pivot-amendment.md).
 
 ## Contributing